SiamHYPER: Learning a Hyperspectral Object Tracker From an RGB-Based Tracker.

Area of Science:

• Computer Vision
• Machine Learning
• Remote Sensing

Background:

• Hyperspectral videos offer rich spatial, spectral, and motion data for tracking camouflaged targets.
• Hyperspectral object tracking faces challenges due to high data dimensionality and the
• data hungry
• problem, hindering model accuracy and generalization.

Purpose of the Study:

• To propose a dual deep Siamese network framework (SiamHYPER) for hyperspectral object tracking.
• To address the
• data hungry
• problem by enabling effective training with limited hyperspectral data.
• To enhance tracking accuracy and generalization by fusing spatial and spectral information.

Main Methods:

• A dual deep Siamese network framework (SiamHYPER) leveraging a pretrained RGB tracker.
• Integration of a hyperspectral target-aware module to mine spectral information.
• Implementation of a spatial-spectral cross-attention module for feature fusion.

Main Results:

• SiamHYPER achieved state-of-the-art performance on a public hyperspectral dataset, outperforming existing trackers.
• Significant improvements in Area Under the Curve (AUC) by 8.9% and 7.2% compared to RGB-based and hyperspectral trackers, respectively.
• High processing speed of 19 FPS, surpassing current hyperspectral tracking methods.

Conclusions:

• The SiamHYPER framework effectively overcomes the
• data hungry
• problem in hyperspectral object tracking.
• The proposed method enables robust and accurate tracking of camouflaged targets using limited training data.
• SiamHYPER offers a promising solution for efficient and high-performance hyperspectral object tracking.